Administrative Information

المعلومات الادارية

عنوان المشروع- Project Title

Colloquial Arabic Speech Recognition Application (CASRA)

كسرة: التعرّف على الكلام العامي المحكي.

الباحث الرئيسي Principal Investigator -

الاسم Name	المؤسسة Institution	الوظيفية Post	العنوان Address	العنوان الالكتروني e-mail	رقم الهاتف Telephone
Ramzi Haraty	Lebanese American University	Assistant Dean, School of Arts and Sciences	P.O. Box 13-5053 Chouran, Beirut, Lebanon 1102 2801	rharaty@lau.edu.lb	01/867620 ext. 1285

الباحثون المشاركون Co-Workers -

لا احد

المدة التعاقدية للمشروع Duration -:

1 year

Scientific Information

المعلومات العلميّة

الهدف- Objectives

The aim is to make the human-computer interaction as natural as possible without prior knowledge of computer skills. Speech enabled computer applications offer this solution. Those applications can perform the required task through the recognition of speech uttered by the user. Speech enabled applications can either be text editors that accepts speech and transform it into text, or can recognize the command uttered by the user and perform it.

ألانجازات المحققة Achievements -

In this study, we have described the development of an automatic speech recognition system based on phoneme-like template model that is dedicated to the Lebanese dialect. The research represented here, to our knowledge, is the first attempt to implement a speech recognition system for the Lebanese colloquial Arabic language. The recognition results produced by our system showed to be satisfactory and when compared, they can match with the results of other Arabic ASR systems. Future work will focus on improving the phoneme-like templates by increasing the size of the training corpus, by trying different distortion measures, calculating 1st and 2nd order delta coefficients and by having multiple codebooks.

آفاق البحث Perspectives -

Although there was, and still continues, extensive research and advancements in speech recognition on English language, there has been little research done on Arabic language. In addition, most of the research done is either for the classical Arabic language or the Egyptian colloquial language. Commercial applications related to this field are mostly based on telephony technology. The research proposed here is for an Arabic speech recognition application, concentrating on the Lebanese dialect. The speech recognition system is middle to a large vocabulary based system that is speaker independent and accepts continuous speech, where the user does not pause between words but speaks naturally. The system starts by sampling the speech, which is transforming the sound from analog to digital, and then extracts the features by using the Mel-Frequency Cepstral Coefficients (MFCC). The extracted feature is compared to the system's stored model; in this case the stored model chosen is a phoneme-based model. Finally, acoustic modeling using Hidden Markov Model (HMM) is applied, were the phonetic model is compared to the system model in order to recognize the word uttered, and the recognized word is sent to the editor.

المنشورات والمساهمات في المؤتمرات- Publications & Communications

1. CASRA+: A Colloquial Arabic Speech Recognition Application. American Journal of Applied Sciences. ISSN: 1546-9239. 4(1): 23-32, 2007.

2. Lebanese Colloquial Arabic Speech Recognition. Proceedings of the ISCA 18th International Conference on Computer Applications in Industry and Engineering (CAINE-2005). Hawaii, USA. November 2005.

موجزعن نتائج البحث Abstract -

The research proposed here is for an Arabic speech recognition application, concentrating on the Lebanese dialect. The system starts by sampling the speech, which is the process of transforming the sound from analog to digital and then extracts the features by using the Mel-Frequency Cepstral Coefficients (MFCC). The extracted features are then compared with the system's stored model; in this case the stored model chosen is a phoneme-based model. This reference model differs from the direct word template matching, where speech features that are extracted from the input are directly compared to the word templates. Each word template in the direct matching model is stored as a vector of feature parameters. Thus, when the vocabulary size of the ASR system becomes large, the memory size for the word template will become humongous. In contrast, the model used here is phoneme-like template matching. Word templates are stored as phoneme-like template parameters. Thus, the memory size for the word templates will not grow as fast as that of the direct matching model.